The foundational world model is here!

The compete for the top spot in AI continues! After OpenAI Sora, Google’s DeepMind team has released a new AI world model called Genie, which contains 11 billion parameters. According to Google, Genie underwent over 200,000 hours of 2D game video training. Google Genie is different from Sora. Genie can generate playable virtual 2D worlds from a single image!

Foundational World Model

Genie is a foundational world model, which means that it can be used to create a variety of different types of environments, including games, simulations, and even educational tools.

Genie models

Genie is trained on a massive dataset of 200,000 hour video game footage. This allows it to learn the physics and rules of different types of environments, as well as how to create characters and objects that can interact with each other.

Genie consists of three parts: a latent action model, used to infer latent actions between each pair of frames; a video tokenizer, used to convert raw video frames into discrete tokens; and a dynamic model, used to predict the next frame of a video given latent actions and past frame tokens.

How to use Genie

To use Genie, users simply need to provide an image of the environment they want to create. Genie will then generate a playable 2D world based on that image. Users can then interact with the world by moving around, jumping, and interacting with objects.

Preliminary Under Development Phase

The current version released is rather primitive, being just an under development research preview version. However, it has the potential to revolutionize the way we create and interact with virtual worlds. It could be used to create new types of games, simulations, and educational tools. It could also be used to help people with disabilities experience the world in a new way.

Here are some of the potential applications of Genie:

• Game development: Genie could be used to create new types of games, including platformers, adventure games, and even RPGs.
• Simulations: Genie could be used to create simulations of real-world environments, such as cities, forests, and even outer space.
• Educational tools: Genie could be used to create educational tools that help students learn about different subjects, such as science, history, and math.
• Assistive technology: Genie could be used to help people with disabilities experience the world in a new way. For example, it could be used to create virtual worlds that people with visual impairments can explore.

Recognized by NVIDIA Jim Fan

Although Genie has been ridiculed by netizens for generating blurry works, it has been praised by industry professionals such as NVIDIA scientist Jim Fan, who stated that, unlike Sora, Genie is actually a world model that can infer actions and is driven by correct actions.